Academic Publications

Predicting Risk in Community Corrections

By
Gabe Haarsma, Sasha Davenport, Devonte C. White, Pablo A. Ormachea, Erin Sheena, David M. Eagleman

Summary

This study introduced a new mobile neuro-cognitive assessment tool to predict recidivism risk among probationers. The tool measures decision-making abilities and traits linked to criminal behavior in an engaging, tablet-based format. Machine learning models were used to quantify recidivism risk. Testing 730 probationers showed the tool has predictive validity comparable to commonly used risk assessments.

Key Takeaways

  •  The neuro-cognitive assessment tool had a predictive validity of 0.70 for recidivism using machine learning models. 
  • This is comparable to the 0.60s-0.70s range seen in widely used risk assessments like COMPAS.
  • The assessment measures dynamic traits like impulsivity and aggression rather than static factors.
  • It is self-administered on a tablet in 30 minutes without specialized training requirements.
  • Eliminating bias and subjectivity while linking rehab to cognitive drivers were key goals.